Tune Java program performance with IBM^® Developer Kit for Java

You should take several aspects of Java™ application performance into consideration when building a Java application for your iSeries™ server.

Here are some links to details and hints on how you can get better performance:

• Improve performance of your Java code by using the Create Java Program (CRTJVAPGM) command, the Just-In-Time compiler, or Using cache for user class loaders.
• Change your optimization levels to achieve the best static compilation performance.
• Carefully set your values for optimal garbage collection performance.
• Only use native methods to start system functions that are relatively long running and are not available directly in Java.
• Use the javac -o option at compilation time to perform method inlining and significantly improve your method call performance.
• Use Java exceptions in cases that are not the normal flow through your application.

Use these tools with the Performance Explorer (PEX) to locate performance problems in your Java programs:

• You can collect Java event trace performance tools using the iSeries Java virtual machine.
• To determine the time that is spent in each Java method, use Java call traces.
• Java profiling locates the relative amount of CPU time that is spent in each Java method and all system functions that are in use by your Java program.
• Use the Java Performance Data Collector to provide profile information about the programs that run on the iSeries server.

Any job session can start and end PEX. Normally, the data that is collected is system wide and pertains to all jobs on the system, including your Java programs. At times, it may be necessary to start and stop the performance collection from inside a Java application. This reduces the collection time and may reduce the large volume of data that is usually produced by a call or return trace. PEX cannot run from within a Java thread. To start and stop a collection, you need to write a native method that communicates to an independent job though a queue or shared memory. Then, the second job starts and stops the collection at the appropriate time.

In addition to application-level performance data, you can use existing iSeries system level performance tools. These tools report statistics on a Java thread basis.

• Java event trace performance tools
  The iSeries Java virtual machine enables the trace of certain Java events.
• Java performance considerations
  Understanding the following considerations can help you improve the performance of your Java applications.
• Java garbage collection
  Garbage collection is the process of freeing storage that is used by objects that are no longer referred to by a program. With garbage collection, programmers no longer have to write error prone code to explicitly "free" or "delete" their objects. This code frequently results in "memory leak" program errors. The garbage collector automatically detects an object or group of objects that the user program can no longer reach. It does this because there are no references to that object in any program structure. Once the object has been collected, you can allocate the space for other uses.
• Java Native Method Invocation performance considerations
  Native method invocation on an iSeries server may not perform as well as native method invocation on other platforms.
• Java method inlining performance considerations
  Method inlining can significantly improve method call performance. Any method that is final is a potential candidate for inlining.
• Java exception performance considerations
  The iSeries exception architecture allows versatile interrupt and retry capabilities. It also allows mixed language interaction. Throwing Java exceptions on an iSeries server may be more expensive than on other platforms. This should not affect overall application performance unless Java exceptions are routinely used in the normal application path.
• Java call trace performance tools
  Java method call traces provide significant performance information about the time that is spent in each Java method.
• Java profiling performance tools
  System wide central processing unit (CPU) profiling calculates the relative amount of CPU time that is spent in each Java method and all system functions in use by your Java program.
• Collect Java performance data
  To collect Java performance data on an iSeries server, follow these steps.